Inclusive finance and biodiversity risk: addressing environmental challenges, improving urbanization landscape resilience and sustainable development

1 Introduction

The 21st century is unfolding as a century of compounding planetary risks. Climate change, biodiversity loss, land-use change, and rapid urbanization now interact in ways that challenge conventional financial and policy responses. The World Economic Forum’s 2025 Global Risks Report ranks biodiversity loss and ecosystem collapse as the second-greatest long-term global threat and estimates that more than half of global GDP is moderately or highly dependent on nature (World Economic Forum, 2025a). Cities sit at the epicentre of this nexus: they generate roughly 70% of global CO2 emissions and are projected to house nearly 70% of the world’s population by 2050 (Franklin, 2024). Financial markets are being reshaped in response. The launch of the Taskforce on Nature-related Financial Disclosures (TNFD) framework, the European Union’s Corporate Sustainability Reporting Directive (CSRD) and Sustainable Finance Disclosure Regulation (SFDR), and the adoption of the Kunming-Montreal Global Biodiversity Framework (KMGBF) in December 2022 have together created a policy architecture in which nature is treated as a material financial concern rather than an externality (The Nature Conservancy, 2025; World Economic Forum, 2025b; Finance for Biodiversity Foundation, 2025). Private nature finance has grown more than tenfold since 2020, but the biodiversity finance gap identified in the KMGBF still stands at approximately US$700 billion per year (The Nature Conservancy, 2025).

The traditional instruments through which nature finance flows-green bonds, sustainability-linked loans, debt-for-nature swaps-are concentrated among large investors and issuers, with limited reach into the small enterprise and household actors whose land-use and consumption choices shape environmental outcomes (World Economic Forum, 2025b). Digital financial inclusion (DFI), the extension of affordable financial services to previously excluded populations through digital platforms, offers a potentially complementary mechanism that operates at the granular scale where those choices are made (Li et al., 2025; Zheng et al., 2025). Recent literature on the digital transformation of urban economies also documents links among digital construction, trade, air quality, and green growth, though findings vary by context and specification (Xu et al., 2026; Wang et al., 2024; Yang W. et al., 2026; Yang L. et al., 2026; Yang et al., 2024). The specific objective of this paper is to test, using a balanced provincial panel from China (2011–2022) and cross-sectional 2026 biodiversity data, whether DFI is empirically associated with reductions in CO2 emissions and lower biodiversity risk scores, and to identify which components of DFI, and which regions of China, drive any observed effects. We treat this as a testable empirical question rather than a presumed positive.

This paper addresses three gaps in the existing literature. First, the biodiversity-finance literature is dominated by review and conceptual work; Jonäll et al. (2025) explicitly identify a shortage of empirical studies connecting specific financial instruments to biodiversity outcomes. Second, the environmental economics literature on DFI has focused overwhelmingly on carbon emissions and aggregate ecological indices (Lu and Xia, 2024; Yu and Wu, 2026; Azizi et al., 2025), typically finding positive environmental effects, but with limited attention to the biodiversity dimension. Third, the urban-resilience literature emphasizes design and governance while under-specifying the financial architecture that determines implementation (Rizwan Ullah et al., 2025; Zhang et al., 2024). This paper contributes by (i) constructing a harmonized provincial panel that links a rigorously measured DFI index (PKU-DFIIC) with the standard national emissions inventory (CEADs) and the WWF Biodiversity Risk Filter, (ii) reporting complete panel regression estimates with province-clustered standard errors, sub-index decompositions, regional heterogeneity, and robustness checks, and (iii) presenting a cross-sectional analysis of DFI–biodiversity relationships that complements the panel-based CO2 analysis.

The distinctive strengths of this paper are threefold. First, it uses only publicly available datasets and reports complete regression tables to ensure full reproducibility. Second, it disaggregates DFI into its three constituent sub-indices (coverage breadth, usage depth, digitization level) and reports separate estimates for each, allowing identification of which delivery channel, if any, drives environmental effects. Third, it distinguishes empirical findings strongly supported by the data from forward-looking policy implications that go beyond observational evidence. The main limitations are the cross-sectional nature of the WWF Biodiversity Risk Filter data (which precludes a panel biodiversity analysis), the exclusion of Tibet from the CEADs provincial inventory, and the absence of a fully specified mediation analysis for channels such as green innovation and industrial upgrading, for which harmonized provincial panels were not available at the time of writing. Section 5.5 documents how these constraints shape the interpretation of the results.

The remainder of the paper proceeds as follows. Section 2 reviews the relevant literature on inclusive finance, biodiversity as financial risk, and urban resilience. Section 3 develops a conceptual framework and states testable hypotheses. Section 4 describes the three data sources: PKU-DFIIC, CEADs, and WWF BRF, and the empirical strategy. Section 5 reports baseline panel estimates, sub-index decomposition, regional heterogeneity, the cross-sectional biodiversity analysis, and robustness checks. Section 6 discusses findings in relation to prior work, including studies of digital urban construction and Chinese regional air quality (Xu et al., 2026; Wang et al., 2026). Section 7 presents a policy roadmap that distinguishes evidence-based near-term measures from forward-looking proposals. Section 8 concludes.

2 Literature review

2.1 Biodiversity as a financial risk

Throughout much of the last century, biodiversity has been viewed, at the very least, as a secondary product of business, secondary to investor value (e.g., a negative externality). This view is no longer sustainable. Azizi et al. (2025) examined the nature of how biodiversity information is being accounted for in non-financial disclosures made by 24 European Financial Institutions and found that biodiversity has begun to shift from the periphery of corporate disclosure to one of increasing importance and centrality due to policies such as the European Taxonomy, Sustainable Finance Disclosure Regulation (SFDR) and Corporate Sustainability Reporting Directive (CSRD). At the Macro Level, 72% of EU companies are believed to rely on ecosystem services in decline, and 75% of loans made by European banks are to these companies. At a Macro Level, many sectors (e.g., fossil fuels, agriculture, fisheries) may be costing the Global Economy between $10 and $25 trillion annually through their impacts on Nature (World Economic Forum, 2025b). Related financial risk research has also moved toward advanced analytical tools, including deep neural network forecasting of financial risk in carbon trading markets (Luo et al., 2024) and large-sample evidence on the risk-return implications of portfolio rebalancing in the Chinese ETF market (Yu and Wang, 2026). These adjacent strands demonstrate how sophisticated financial risk analytics have become, while biodiversity-related financial risk remains comparatively underdeveloped.

Two recent reviews were particularly valuable in identifying opportunities for future work. The first review, by Jonäll et al. (2025), synthesizes literature on biodiversity and finance, and identifies five key under-represented areas for research: (1) Biodiversity Strategies in Corporations and Finance, (2) Measurement and Reporting of Biodiversity in Corporations and Finance (3) Stakeholder and Sectoral Engagement; (4) Foundational Approaches to Actionable Solutions; and (5) Empirical Case Studies. The second review, by Azizi et al. (2025), examines how European Financial Institutions disclose biodiversity information and concludes that they are still in the early stages of developing their biodiversity reporting standards, with significant variation in the quality and scope of that reporting. According to Azizi et al. (2025), Institutional Theory can help explain how the increasing amount of regulation (CSRD and SFDR) will act as a catalyst for shaping a sustainable transition in the Financial Sector. Taken together, these approaches to biodiversity and finance highlight that biodiversity is an important transitional risk for financial institutions (due to growing regulatory and reputational risks) as well as a significant physical risk (resulting from the degradation of ecosystem services). This transition will have profound implications for the financial services industries (banks and insurers) whose portfolios are significantly dependent on agricultural yields, water availability and coastal stability.

2.2 Inclusive finance and environmental outcomes

The extensive body of empirical research on the impacts of financial development on the environment has been expanding rapidly, with much of the early work examining the extent to which financial development is beneficial or harmful to environmental quality, and results heavily influenced by governance and technology diffusion (Shashwat et al., 2025; Rizwan Ullah et al., 2025). The rapidly growing area of digital financial inclusion (DFI) offers a distinct new means of advancing environmental sustainability by reducing transaction costs, expanding coverage for low-income and rural economic segments, and embedding financial services into digital platforms that also provide information and promote green consumption (Ye et al., 2023; Xiong et al., 2022). Evidence from China’s health poverty alleviation programme also shows that targeted financial risk protection can materially strengthen the resilience of vulnerable rural households to illness-related financial shocks (Chen and Pan, 2019). This supports the broader capability argument that access to protective financial mechanisms can improve household resilience, even though the intervention is not itself a DFI programme.

In the case of China, the empirical evidence is especially comprehensive (Li, 2024). Lu and Xia (2024), based on panel data from 276 prefecture-level cities over the period 2011–2020, demonstrated that DFI is responsible for reducing carbon emissions in a statistically significant way and that DFI leads to positive spatial spillovers with respect to other cities and regions, with approximately 59% of the effect mediated through green technological innovation (Lu and Xia, 2024; Ding et al., 2022). Li et al. (2025) reported a 3.89% increase in a composite ecological-environment index attributable to DFI, using a panel of 733 counties within the Yellow River Basin during 2014–2022, with technological innovation and structural rationalization of industries as the primary transmission mechanisms. Through their examination of DFI’s impact on counties located within National Key Ecological Functional Zones (Mo et al., 2025; Yu and Wu, 2026) demonstrated that DFI fosters growth in the green economy, through advances in agricultural technology, upgrades to the industrial structure and innovation by businesses; however, the effects of DFI are weaker in western and northeastern areas of China, as well as in areas designated as windbreak-sand-fixing zones. Finally, Zheng et al. (2024) have used a dual machine-learning framework to extend these authors’ findings and identify the causal impact of DFI on the intensity of agricultural carbon emissions. Recent computational and quasi-experimental studies further broaden this evidence base. Xia et al. use exploratory spatial data analysis and graph convolutional neural networks to capture regional dependence in the evolution of DFI across China (Xia et al., 2025). Jiang and Yuan, using a staggered difference-in-differences design across Chinese cities, find that the e-CNY pilot is associated with consumption upgrading, with regional development and fintech depth shaping the magnitude of the effect (Jiang and Yuan, 2026).

Outside China, similar mechanisms have been documented in other countries. Ullah et al. (2025), in a study of 25 OECD Countries (1991–2022), found that a combination of ICT investments, Financial Inclusion and Good Governance supports improvements in Environmental Performance. Shashwat et al. (2025) researched middle- and upper-middle-income countries and found that Digital Financial Inclusion (DFI) enhances Green Growth, depending on the quality of governance and the level of Economic Globalization. Basnayake et al. (2024) write that in an analysis of 30 Asia-Pacific countries, there are Non-Linear Threshold Effects of DFI on Economic Development (Basnayake et al., 2024; Hansen, 1999). These studies convey the same message: Digital Inclusive Finance supports environmental benefits, but only when accompanied by a Joint Presence of Complementary Partnerships across Institutions, Technology, and Regulations.

2.3 Urbanization landscape resilience

The connection between urbanization, environmental quality and biodiversity manifests itself in the design of the urban landscape most obviously through Urban Green Infrastructure (UGI), i.e., parks, street trees, green roofs, wetlands and blue–green corridors. Urban Green Infrastructure (UGI) has multiple ecosystem services, including regulating, provisioning, cultural, and supporting services. UGI has been shown, through systematic reviews, to mitigate urban heat, manage stormwater, enhance quality of life by improving mental health and social ties, and promote biodiversity conservation (Sokolova et al., 2024). Research has demonstrated that designs that provide multiple functions by integrating engineered elements, such as bioswales and green roofs with natural elements such as urban forests and wetlands allow for maximum performance of UGI; however, as Cook et al. (2024) state, if too many designs are being analyzed simultaneously without a well-defined criterion or matrix for evaluation, it can create a paradox of choice that hinders implementation. Recent work on China’s resource-based cities likewise shows that green transition resilience is strongly spatially heterogeneous and is associated with government intervention, income, human capital, openness, and environmental regulation (Wang et al., 2026). For urban wetlands, integrated conservation-restoration planning has also been shown to improve multiple ecosystem services, including carbon storage, flood regulation, habitat quality, and landscape and cultural benefits, while explicitly accounting for budget constraints (Li et al., 2026).

The challenge of implementing urban resilience will also be magnified by the financial commitment needed to meet the projected demands of an urbanizing population. “An estimated 70% of the world’s population will live in cities” by 2050, which will increase the demand for housing, energy, water and mobility infrastructure (One More Tree Foundation, 2025; Franklin, 2024). Cities like Copenhagen, Singapore and Rotterdam have established comprehensive climate adaptation strategies that integrate retention parks, ecological corridors and coastal protection, yet, as Wang et al. (2024) point out, the limiting factor to implementation is typically not a lack of good design ideas, but rather the combination of funds, governance, and equity necessary to implement good design at scale (Song and Du, 2026). This is where urban landscape literature connects with the finance literature. At the broader social-ecological scale, network analysis of coastal wetland systems shows that ecologically critical nodes and stakeholder attention can diverge, making the alignment of ecological priorities and social acceptance an important governance problem (Zhi et al., 2026). A knowledge-data-driven hybrid model applied to 271 counties in the Central Plains Urban Agglomeration further identifies nonlinear and spatially heterogeneous effects of social and ecological pressures on urban resilience (Wu, Li & Lin, 2026). Research on smart transportation technologies similarly cautions that digital urban innovation can improve social welfare while creating accumulating environmental burdens, so digitalization should not be presumed to be environmentally benign (Xu et al., 2026).

2.4 Research gaps

There are still three gaps. Firstly, at a theoretical and descriptive level, most of the literature on the biodiversity–finance nexus focuses on theory and description, with less empirical evidence linking financial instruments to sub-national biodiversity outcomes (Jonäll et al., 2025). Secondly, the environmental economics literature on DFI often overemphasises carbon emissions and pollution aggregates, at the expense of addressing specific details of biodiversity and urban landscapes (Lu and Xia, 2024). Finally, much of the research into urban resilience has concentrated on issues of design, governance, and equity, whilst failing to provide comprehensive integration of the financial structures that support the implementation of urban resilience. This paper will provide an integrated framework for all three of these areas of research.

3 Conceptual framework and hypotheses

Using the three bodies of literature reviewed above as a foundation, we have constructed a conceptual framework (illustrated in Figure 1) that considers inclusive finance as an enabling force influencing both the biodiversity risk of Urbanization and the resilience of the Urbanization Landscape through three mediating mechanisms. These mechanisms are moderated by both the institutional and ecological context.

The first mediating mechanism relates to the development of Green Technological Innovation. Through Inclusive Finance, small-to medium-sized enterprises and innovative start-ups can now access sufficient capital to invest in clean technologies, resource-efficient production, and biodiversity-related products (Lu and Xia, 2024; Tan et al., 2025). The second mechanism relates to the upgrading of Industrial Structures. By shifting financial resources from resource-intensive industries to Service and Advanced Manufacturing, DFIs reduce pollution intensity and create new jobs and activities that are less environmentally destructive (Li et al., 2025; Yu and Wu, 2026). The third mediating mechanism deals with capability. Access to additional types of credit, insurance, savings, and digital payments gives Micro and Small- and Medium-Sized Enterprises (MSMEs) greater capacity to adopt clean energy sources, mitigate climate and weather risks, and participate in Green Value Chains (Yan et al., 2024; Liu et al., 2025; Li et al., 2025). Organizational-network evidence provides a complementary mechanism: strong and weak ties can influence business-model innovation through knowledge sharing, while knowledge integration conditions the strength of that pathway (Wu et al., 2026). This supports treating knowledge capability as an important boundary condition around the green-innovation mechanism proposed here.

The Mechanisms for Biodiversity Risk Mitigation and Resilience in Urban Landscapes are Diverse, and each produces two distinct outcomes. Biodiversity risk mitigation reduces pressure on species and ecosystems from the negative impacts of land-use change, pollution, and over-extraction of natural resources. The ability of urban landscapes to withstand and recover from stressors (e.g., disturbance) and continue to provide ecosystem services (e.g., remain functional) after those stressors is defined as the “resilience” of urban landscapes. Both the mechanisms of Biodiversity Risk Mitigation and Urban Landscape Resilience interact, as Biodiverse Urban Landscapes produce more resilient Ecosystem Services that support Biodiversity by reducing fragmentation and edge effects.

While the Mechanisms for Biodiversity Risk Mitigation and Resilience in Urban Landscapes are interconnected, it is important to note that each has unique effects that may vary across contexts. We have identified four factors that moderate the interaction between these two mechanisms: Regulatory Standards/Environmental Regulation increases the marginal utility of Clean Investment and holds financial markets more accountable for the use of sustainable financial products. Disclosure Standards (e.g., TNFD) hold nature-damaging activities to much higher standards than they do their direct environmental impact. Urban Green Infrastructure provides the physical framework within which financial capital translates to the resilience of Urban Landscapes. Finally, Governance Quality (Regulatory Capacity, Rule of Law and Control of Corruption) shapes the influences of all three (Ullah et al., 2025).

From this framework, we derive four testable hypotheses.

H1 Higher inclusive finance intensity is associated with reduced biodiversity risk and improved ecological and environmental outcomes at the sub-national level.

H2 Inclusive finance strengthens urbanization landscape resilience, measured by a composite index of urban green cover, infrastructure quality, and ecosystem-service provision.

H3 The effects of inclusive finance on environmental and urban resilience outcomes are mediated by green technological innovation, industrial structure upgrading, and household/MSME capabilities.

H4 The effects are moderated by environmental regulation, disclosure standards, urban green infrastructure, and governance quality, with stronger impacts where these conditions are more developed.

4 Data and methodology

4.1 Data sources

We construct a balanced provincial panel covering 30 mainland Chinese provinces, municipalities and autonomous regions over 2011–2022, yielding 360 province–year observations. Tibet is excluded because the CEADs provincial CO2 emission inventory does not report Tibet, following the standard convention in the Chinese emissions literature. Digital financial inclusion is measured using the Peking University Digital Financial Inclusion Index of China (PKU-DFIIC), developed by the Institute of Digital Finance at Peking University in collaboration with Ant Group and constructed from digital transaction records aggregated to the provincial level (Sokolova et al., 2024). The index reports an aggregate score on a 0–500 scale together with three sub-indices: coverage breadth (number of accounts, cards, and access points per capita), usage depth (payment, credit, insurance, investment and monetary-fund activity per user), and digitization level (mobile-based transaction share, credit availability, and pricing accessibility). We use the 2011–2023 wave and retain the full 2011–2022 window that overlaps with our emissions data.

The outcome variable in the panel model is provincial apparent CO2 emissions, in millions of tonnes, obtained from the CEADs (China Emission Accounts and Datasets) provincial apparent CO2 emission inventory for 1997–2022 [maintained by researchers at UCL, Tsinghua, and other institutions] (Table 1). The apparent-emissions accounting method sums CO2 emissions from 17 fossil-fuel types and cement production, using province-specific emission factors, and is the standard measure used in Chinese sub-national emissions research. For the biodiversity analysis, we use the WWF Biodiversity Risk Filter (BRF) v3.0 sub-national risk scores released in 2026, downloaded via the country-profile download function at riskfilter.org. The BRF is a cross-sectional composite that aggregates 70 indicators across three domains: physical risk (12 provisioning, regulating, and cultural service categories), regulatory-deficiency risk (uniform at the national level for all Chinese provinces at 2.75), and reputational risk (13 pressure and environmental factor sub-scores). Because the BRF is a single 2026 cross-section, we do not treat it as a panel outcome; instead, we relate province-level 12-year mean DFI to province-level BRF scores in a cross-sectional specification (Section 5.4). This is a substantive design choice and a real constraint of the biodiversity data currently available at the sub-national scale. We follow the broader methodological principle of transparent indicator construction and explicit reporting of spatial driver analysis, as emphasized in recent sub-national environmental research (e.g., the Geodetector-based approach in Wang et al. (2025), even though that study addresses a different outcome (vegetation carbon sinks) and region.

TABLE 1. Variable definitions and sources.

The panel sample comprises 30 provinces × 12 years (N = 360). Cross-sectional biodiversity analysis uses N = 30 provinces with 2011–2022 mean DFI, matched to 2026 BRF, scores; CEADs, China Emission Accounts and Datasets; PKU-DFIIC, peking university digital financial inclusion index; WWF BRF = WWF, Biodiversity Risk Filter (v3.0, 2026).

4.2 Empirical strategy

To test the DFI–emissions relationship, we estimate the following two-way fixed-effects panel model for province i in year t:

ln CO 2 i t = α + β · ⁡ ln DFI i t + μ i + λ t + ε i t (1)

Where ln CO 2 i t is the natural log of provincial apparent CO2 emissions in millions of tonnes, ln DFI i t is the natural log of the PKU-DFIIC aggregate index, μ i are province fixed effects that absorb all time-invariant provincial characteristics (geography, initial industrial structure, endowments), λ t are year fixed effects that absorb common national shocks (macroeconomic cycles, national policy, COVID-19), and ε i t is the error term. Standard errors are clustered at the province level to allow for arbitrary within-province correlation over time. The coefficient β identifies the elasticity of CO2 emissions with respect to DFI, purged of time-invariant province heterogeneity and common year shocks.

We supplement the baseline model with three additional specifications. First, we decompose DFI into its three sub-indices (coverage breadth, usage depth, digitization level) and re-estimate Equation 1 with each sub-index alone and with all three jointly, to identify which delivery channel drives any aggregate effect. Second, we estimate Equation 1 separately for each of China’s four traditional macro-regions (East, Central, West, Northeast) to examine regional heterogeneity. Third, for the biodiversity analysis, we estimate cross-sectional OLS regressions of the form BRF i = α + β · ln (mean DFI i) + ε i for each of nine BRF risk categories, using HC1-robust standard errors and N = 30 provinces. Because BRF is a single cross-section, this specification identifies associations rather than causal effects. We did not implement the province–year internet-penetration × Alipay-rollout instrument proposed in earlier Chinese DFI research, because internet penetration plausibly influences emissions and biodiversity through channels independent of DFI (including green innovation, industrial digitization, and information-based environmental governance), violating the exclusion restriction required for a valid instrument. We therefore rely on fixed-effects identification, augmented by (i) one- and two-year lagged DFI specifications that reduce simultaneity bias, and (ii) inclusion of province-specific linear time trends that absorb slow-moving unobserved province trajectories. We characterize the resulting estimates as associations under strong conditional independence assumptions.

Robustness checks include: (i) sub-sampling to the pre-COVID period (2011–2019); (ii) using one-year and two-year lagged DFI as regressors; (iii) estimating in levels rather than logs; and (iv) adding province-specific linear time trends. Section 5.4 reports these results.

5 Results

5.1 Descriptive patterns

Table 2 reports descriptive statistics for the panel. The PKU-DFIIC aggregate score ranges from 18.3 (Guizhou, 2011) to 460.7 (Shanghai, 2022), with a sample mean of 243.9. CO2 emissions range from 41.3 Mt (Qinghai, 2011) to 2,247.7 Mt (Shandong, 2022), with a mean of 391.3 Mt. Biodiversity Risk Filter physical risk scores for Chinese provinces range from 2.74 to 3.73 (mean 3.24, on a 1–5 scale), and reputational risk scores from 3.27 to 4.06 (mean 3.72). All Chinese provinces receive an identical BRF regulatory-deficiency score of 2.75 because that sub-index is derived from national biodiversity laws and strategies applied uniformly across provinces; we therefore drop it from the outcome set.

TABLE 2. Descriptive statistics for the merged panel (30 provinces × 12 years).

Figure 2 shows the global context. Private nature finance rose from US$9.4 billion in 2020 to over US$102 billion in 2024, but the annual biodiversity finance gap identified in the KMGBF is estimated at approximately US$700 billion per year, essentially unchanged over the period (World Economic Forum, 2025a; The Nature Conservancy, 2025). The economic case in panel (b) juxtaposes the estimated US$25 trillion annual cost of nature-related impacts from fossil fuels, agriculture and fisheries [1] against a US$2.7 trillion annual investment need and US$10.1 trillion annual business opportunity from a nature-positive transition.

Figure 3 shows the growth of the PKU-DFIIC across representative provinces from 2011 to 2022. All provinces exhibit steep S-shaped growth, with the national mean rising from 40.0 in 2011 to 393.7 in 2023. Variance decomposition (Section 5.5) shows that 94% of the variation in DFI in our sample is within-province time variation, and only 6% is between-province, meaning most of the DFI signal in our panel is the common national growth trend.

5.2 Baseline estimates

Table 3 reports the baseline estimates. Column (1) shows a simple pooled OLS of ln (CO2) on ln (DFI) with province-clustered standard errors: the coefficient is 0.038 (SE = 0.053, p = 0.47) and statistically indistinguishable from zero. Column (2) adds province fixed effects, yielding a marginally significant positive coefficient (β = 0.049, SE = 0.027, p = 0.068). Column (3), our main specification, adds year fixed effects to obtain a two-way fixed-effects estimate: the coefficient is 0.047 (SE = 0.124, p = 0.70), statistically indistinguishable from zero. Column (4) further adds province-specific linear time trends and yields essentially zero (β = 0.002, p = 0.95). The consistent message is that once province and year fixed effects absorb the common growth trends in both DFI and emissions, no robust effect of aggregate DFI on CO2 emissions remains. Figure 4 presents these estimates graphically with 95% confidence intervals.

TABLE 3. Baseline two-way fixed-effects panel estimates: ln (CO2) on ln (DFI).

This null result on the aggregate index does not, however, imply that DFI is unrelated to environmental outcomes. Table 4 decomposes DFI into its three sub-indices. In univariate specifications, coverage breadth (β = 0.015, p = 0.71) and usage depth (β = 0.085, p = 0.47) are not significant, but the digitization sub-index enters negatively and is marginally significant (β = −0.163, p = 0.057). In the joint specification that includes all three sub-indices simultaneously, digitization becomes significantly negative (β = −0.190, SE = 0.091, p = 0.038). This suggests that the channel through which DFI can plausibly reduce emissions is the digital delivery component: the substitution of physical financial infrastructure and paper-based transactions with mobile and online channels, rather than sheer expansion of coverage.

TABLE 4. DFI sub-index decomposition (two-way fixed effects, ln (CO2) outcome).

5.3 Heterogeneity

Table 5 reports regional two-way fixed-effects estimates by macro-region. In the Eastern coastal region (10 provinces, N = 120), the DFI coefficient is positive but not significant (β = 0.230, p = 0.47). In the Central region (6 provinces, N = 72), the estimate is negative but noisy (β = −1.129, p = 0.52). In the Western region (11 provinces, N = 132), the effect is close to zero (β = −0.101, p = 0.74). The one region with a robust and highly significant negative effect is the Northeast (3 provinces; Liaoning, Jilin, Heilongjiang N = 36), where the DFI coefficient is −0.782 (SE = 0.094, p < 0.001). Figure 5 displays these estimates with 95% confidence intervals. This regional pattern is consistent with the interpretation that DFI growth in the Northeast has coincided with the region’s structural industrial decline, a period in which heavy-industry facilities have been closed or relocated, delivering large emissions reductions concurrently with (but not necessarily caused by) DFI expansion. We are careful not to interpret this as a treatment effect of DFI alone.

TABLE 5. Regional heterogeneity: two-way FE estimates by macro-region.

5.4 Robustness

Table 6 reports robustness checks. Restricting the sample to the pre-COVID period (2011–2019) yields β = 0.085 (p = 0.36), still statistically indistinguishable from zero. Using one-year and two-year lagged DFI as the regressor yields β = −0.042 (p = 0.75) and β = −0.088 (p = 0.53), respectively; negative point estimates but not statistically significant. Level-level estimation yields β = −1.66 (p = 0.17). Including province-specific linear time trends yields β = 0.002 (p = 0.95). Across all specifications, no significant aggregate DFI effect on CO2 emissions emerges. The variance-decomposition result shows that 94% of DFI variation is within-province time variation, and only 6% is between-province, which helps explain why: after year fixed effects absorb the common national DFI trajectory, the residual within-province variation is modest, giving the fixed-effects estimator limited power. This is a genuine feature of the identification problem, not a defect of the estimator.

TABLE 6. Robustness checks: alternative panel specifications (ln (CO2) outcome).

5.5 Cross-sectional biodiversity analysis

Table 7 reports the cross-sectional biodiversity results. Provinces with higher mean DFI show significantly higher BRF scores on reputational risk (β = 0.836, p = 0.005), regulating-services categories (both enabling and mitigating, β ≈ 3.0–3.3, p < 0.001), pressures on biodiversity (β = 1.534, p = 0.003), and socioeconomic factors (β = 0.797, p < 0.001).

TABLE 7. Cross-sectional OLS: mean ln (DFI) 2011–2022 → BRF 2026 category scores (N = 30).

6 Discussion

The results reshape the empirical picture of DFI’s environmental role in three ways. First, contrary to the more optimistic conclusions of recent single-outcome studies focused on carbon emissions in specific Chinese sub-regions (Li et al., 2025; Lu and Xia, 2024; Azizi et al., 2025), we find no robust aggregate DFI effect on provincial CO2 emissions in a balanced 2011–2022 panel. This does not overturn the earlier literature; those studies use different samples, different outcome measures, and often longer or shorter time windows, but it does indicate that the environmental effect of DFI is more contingent than headline claims suggest (Chen, 2026). Second, the sub-index decomposition offers a mechanistically sharper reading: digitization of financial services-the shift from physical to mobile and online delivery-is the channel with the strongest (though still marginal) negative effect on emissions, whereas coverage breadth and usage depth are not significant. This finding is complementary to the digital-urban-construction literature (Xu et al., 2026), which documents positive economic effects of urban digital infrastructure; our result suggests that the environmental dividends of digital finance flow specifically through the digital delivery channel. Third, the strongly negative estimate in the Northeast is consistent with recent evidence on regional air-quality improvement in the same region (Wang et al., 2026), though we attribute the emissions decline primarily to structural industrial change coinciding with DFI growth rather than to DFI itself.

The cross-sectional biodiversity results (Table 7; Section 5.5) require careful interpretation. At first reading, the positive coefficients look like DFI is worsening biodiversity outcomes. The more accurate interpretation is a development-conditional one: high-DFI provinces- Beijing, Shanghai, Zhejiang, Jiangsu, Guangdong are also the most economically developed, densely urbanized, and industrially intense, and it is these characteristics that drive high biodiversity risk. DFI is correlated with development, and development brings biodiversity pressure. Figure 6 makes this clear: the positive relationship is evident, and so is the fact that low-DFI provinces (Qinghai, Xinjiang, Ningxia) have lower biodiversity pressure. The policy implication is that DFI does not automatically deliver biodiversity co-benefits; complementary conservation policy is required.

These findings relate to broader debates about the digital transformation of the economy and its uneven environmental consequences. Our results show that positive-sounding narratives about digital finance and nature need to be tested carefully: the answers differ by outcome, by DFI component, and by region. The DFI–environment relationship is real but conditional, not automatic.

7 Policy implications: a three-horizon roadmap

We distinguish here between two categories of policy implications: (i) implications directly supported by our provincial-panel evidence, and (ii) forward-looking proposals that are consistent with our framework but require additional evidence for full support.

Implications directly supported by the evidence: (1) Because the environmental effect of DFI operates primarily through the digitization sub-index, policies that accelerate the digital delivery of financial services-mobile payments, online credit, digital identity for the unbanked-carry stronger environmental co-benefits than measures that simply expand coverage without altering delivery mode. (2) Because high-DFI provinces carry high biodiversity risk driven by their development intensity, DFI expansion in developed provinces should be complemented by biodiversity-specific policies (protected-area expansion, ecological corridors, urban green infrastructure) rather than treated as a substitute. (3) Because effects are concentrated in the Northeast and coincide with structural industrial change, the Northeast case cannot be generalized as a template for other regions without careful assessment of the underlying industrial dynamics.

Forward-looking policy proposals, including biodiversity risk filters in fintech platforms, green-credit quotas for MSMEs, TNFD/CSRD disclosure integration with DFI data infrastructure, urban green-infrastructure bonds, blended finance for nature-based urban infrastructure, and incorporation of biodiversity into credit ratings and capital adequacy, go beyond what the present provincial observational analysis can directly support. We include them in Figure 7 as a three-horizon roadmap because they align with the KMGBF, TNFD, and CSRD architectures and with the emerging biodiversity-finance literature (World Economic Forum, 2025; Finance for Biodiversity Foundation, 2025; Basnayake et al., 2024), but we clearly flag that the empirical case for each requires further study, including instrument-specific impact evaluations.

The long-horizon suggestion of embedding biodiversity into credit ratings and capital adequacy is, in this paper, a proposal drawn from the analogy to climate risk being integrated into financial supervision globally; it is not a finding of our estimation. We include it to indicate a plausible direction of financial system reform, not to claim empirical evidence for its effectiveness. A properly designed evaluation would require post-implementation data from jurisdictions that adopt such measures.

8 Conclusion

This paper has re-examined the relationship between digital financial inclusion, carbon emissions, and biodiversity risk in China using a balanced provincial panel of 30 provinces over 2011–2022, and a cross-sectional biodiversity analysis of the same 30 provinces using WWF Biodiversity Risk Filter 2026 scores. Contrary to the prevailing narrative that digital financial inclusion is uniformly nature-positive, our two-way fixed-effects estimates show no statistically significant effect of the aggregate DFI index on provincial CO2 emissions. The environmental benefit of DFI, where it exists, comes from the digitisation sub-index and is concentrated in the Northeast region; a finding that plausibly reflects the co-occurrence of DFI growth with regional industrial restructuring rather than a DFI treatment effect. Cross-sectionally, high-DFI provinces have higher biodiversity risk scores because they are also the most economically developed and urbanised, meaning DFI does not automatically deliver biodiversity co-benefits at the provincial scale.

Three limitations frame the future-research agenda. First, the WWF Biodiversity Risk Filter is a 2026 cross-section, which precludes a proper panel biodiversity analysis; as sub-national time-series biodiversity indicators develop, our cross-sectional findings should be re-examined dynamically. Second, we could not obtain harmonised provincial panels for the mediating channels (green innovation, industrial structure upgrading, household/MSME capability) at the time of writing; therefore, we do not report a formal mediation decomposition. A full mediation analysis using harmonised provincial data on these channels is the highest-priority extension. Third, the paper focuses on China’s 30 mainland provinces (Tibet excluded); cross-country comparative analysis, especially in sub-Saharan Africa, Southeast Asia, and Latin America, would test the external validity of the mechanisms identified here. These limitations notwithstanding, the paper provides transparent, reproducible provincial evidence and clarifies both what digital financial inclusion appears to do for the environment and what it does not.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Xiwei Shen, University of Nevada, United States

Reviewed by: Hu Li, Sichuan University of Science and Engineering, China

Yuxun Gu, Xi’an Jiaotong-Liverpool University, China